An Estimation of the Optimal Gaussian Kernel Parameter for Support Vector Classification

Authors:
Wenjian Wang;Liang Ma
Affiliations:
School of Computer and Information Technology Key Laboratory of Computational Intelligence & Chinese Information Processing of Ministry of Education, Shanxi University, Taiyuan, P.R.C 030006;School of Computer and Information Technology Key Laboratory of Computational Intelligence & Chinese Information Processing of Ministry of Education, Shanxi University, Taiyuan, P.R.C 030006
Venue:
ISNN '08 Proceedings of the 5th international symposium on Neural Networks: Advances in Neural Networks
Year:
2008

Citing 6
Cited 0

The nature of statistical learning theory

The nature of statistical learning theory
Clustering by Scale-Space Filtering

IEEE Transactions on Pattern Analysis and Machine Intelligence
Conformal Transformation of Kernel Functions: A Data-Dependent Way to Improve Support Vector Machine Classifiers

Neural Processing Letters
Learning to Predict the Leave-One-Out Error of Kernel Based Classifiers

ICANN '01 Proceedings of the International Conference on Artificial Neural Networks
Applications of Support Vector Machines for Pattern Recognition: A Survey

SVM '02 Proceedings of the First International Workshop on Pattern Recognition with Support Vector Machines
Financial time series prediction using least squares support vector machines within the evidence framework

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.00

Visualization

Abstract

The selection of kernel function and its parameters has heavy influence on the generalization performance of support vector machine (SVM), and it becomes a focus on SVM researches. At present, there are not general rules to select an optimal kernel function for a given problem yet, alternatively, Gaussian and Polynomial kernels are commonly used for practice applications. Based on the relationship analysis of Gaussian kernel support vector machine and scale space theory, this paper proves the existence of a certain range of the parameter 茂戮驴, within the range the generalization performance is good. An appropriate 茂戮驴within the range can be achieved via dynamic evaluation as well. Simulation results demonstrate the feasibility and effectiveness of the presented approach.